JP4394083B2 - Signal detection apparatus, signal detection method, signal detection program, and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a signal detection apparatus for detecting a signal at high speed, in which a search signal is extracted from an input signal at coarse spacing, and a similar signal with the search signal is selected from an accumulated signal DB (data base) by the small number of search signals, and by the DB which collects this, the similar signal is searched with the search signal extracted from the input signal at dense spacing. <P>SOLUTION: The signal detection apparatus of the invention comprises: the accumulated signal DB for registering an accumulated signal; an upper search signal extraction section for extracting an upper search signal from a plurality of places of the input signal; an upper search section for searching the accumulated signal in which a similar period is included in the upper search signal is searched from the accumulated signal DB; a lower accumulated signal DB creation section for creating the lower accumulated signal DB in which the accumulated signal in the last search is collected; a lower search signal extraction section for extracting the lower search key signal from the plurality of places of the input signal; and the lower search section for searching the accumulated signal in which the similar period is included in the lower search signal. Therefore, searching is performed according to the number of stages. <P>COPYRIGHT: (C)2008,JPO&amp;INPIT

Description

The present invention relates to a signal search technique for searching for a signal similar to an input signal to be input, and particularly includes a signal similar to a search key signal extracted from an input signal from a stored signal database in which the signal is stored. The present invention relates to a signal detection device, a signal detection method, a signal detection program, and a recording medium that detect a section (signal detection).
For example, the signal search technique is used for detection and list creation of music and video clips used in television broadcasting, or music detection and list creation used in radio broadcasting.

  As the conventional signal search technique, when a specific signal (hereinafter referred to as a reference signal) is given, a reference signal is included from a plurality of signals stored in a database (hereinafter referred to as a stored signal). A method of detecting an interval is used. As this detection method, a time-series active search method (for example, see Patent Document 1, TAS), a division match search method (for example, see Patent Document 2, DAL-1), a faster division match search method (for example, Patent Document 1) 3, DAL-2), a binary area matching method (for example, Patent Document 4, BAM) has been proposed.

  When an input signal is given to a search device having an accumulated signal database in which a plurality of accumulated signals are registered, search key signals are sequentially cut out from the input signals, and the cut-out search key signals are referred to as reference signals. As described above, an accumulated signal having a section similar to a certain section in the accumulated signal in the accumulated signal database is searched. As a result, when the accumulated signal registered in the accumulated signal database is used at the location where the search key signal for the input signal is cut out, the accumulated signal is used at the location where the search key signal for the input signal is cut out. It can be detected as being used.

As described above, for example, an input signal is an audio signal for television broadcasting, and a search key signal having a length of 5 seconds is sequentially extracted from the input signal while being shifted by 1 second, and is included in each extracted search key signal. The stored music is searched from a stored signal database using a music acoustic signal as a stored signal. As a result of this search, it is possible to create a playlist indicating from which time of the input signal the music registered in the stored signal database was used. In addition, by using the video signal, it is possible to create a video content playlist by the same processing as in the case of the music described above.
Note that the fineness of the time information of the playlist depends on the search key signal cut-out interval, and the finer the search key signal cut-out interval, the finer the playlist time information can be detected.
Japanese Patent No. 30653314 JP 2004-102023 A WO2006 / 004050 WO2006 / 006528

Conventionally, it is conceivable to perform signal detection as described above and create a playlist such as music content or video content.
However, in the above conventional example, as the search key signal cut-out interval decreases, the number of search key signals increases, that is, the number of times search processing using the search key increases.
As a result, in the conventional example, when the time information of the playlist is to be detected in detail, there is a problem that it takes a long time to detect the signal in order to reduce the cut-out interval of the search key signal.

Similarly, in the conventional example, as the number of accumulated signals registered in the accumulated signal database increases, it takes for each search key signal to be searched from the accumulated signal database for similar signals. There is a problem that the time increases and the time required for signal detection increases.
The present invention has been made in view of such circumstances, and the search key signal is cut out from the input signal at rough intervals, and is similar to each search key signal from the stored signal database with a small number of search key signals. A signal detection device that performs signal detection at a higher speed by selecting a stored signal and searching for a similar signal with respect to a search key signal extracted from the input signal at close intervals from a database in which these small number of stored signals are collected. An object of the present invention is to provide a signal detection method, a signal detection program, and a recording medium.

The signal detection apparatus according to the present invention compares a predetermined section of an input signal with a part or all of the stored signals registered in advance, and detects a stored signal similar to the input signal at a plurality of search stages. An accumulated signal database in which accumulated signals are registered,
An upper search key signal cutout unit that cuts out an upper search key signal from the input signal at predetermined intervals, and an accumulation signal including a section similar to the upper search key signal is searched from the accumulated signal database and searched. A high-order search unit that outputs the stored signal that has been stored;
And the lower accumulation signal database creation unit that creates a lower accumulation signal database a collection of accumulated signals the upper search unit has output, the lower search key signal at intervals narrower than the spacing in the upper search key signal extracting unit, wherein Similar to the lower search key signal, and a lower search key signal cutout unit that cuts out only from a region in the vicinity of the cutout location of the upper search key signal in which the stored signal including a similar section is searched for in the upper search unit in the input signal A low-order search unit that searches the low-order stored signal database for an accumulated signal including a section to be stored.

The signal detection method of the present invention compares a predetermined interval of an input signal with an interval of an accumulated signal registered in advance in an accumulated signal database, and detects accumulated signals similar to the input signal in a plurality of search stages. The plurality of search stages include one coarse search stage and one or more dense search stages that are sequentially executed following the coarse search stage. The coarse search stage includes: An upper search key signal cutout process in which the upper search key signal cutout section cuts out the upper search key signal from a plurality of locations of the input signal at predetermined intervals, and a section in which the upper search section is similar to the upper search key signal accumulation signal including said searched from stored signal database, wherein the upper search process for outputting the searched stored signal, wherein the dense search step, portion creates lower accumulation signal database And the lower accumulation signal database creation process of creating a lower accumulation signal database a collection of accumulated signals searched by the search for search stage that is performed immediately before said seal search stage, the lower search key signal extracting unit is lower search key signal If, in the input signal, the search stage executed immediately before is a coarse search stage, an area in the vicinity of the cut-out location of the upper search key signal in which the stored signal including a similar section is searched in the coarse search stage Only if the search stage executed immediately before is a dense search stage, only the region in the vicinity of the cut-out position of the lower search key signal in which the accumulated signal including the similar section was searched in the previous dense search stage is searched. from a lower search key signal clipping process of cutting out at the dense search stage narrower than cut interval of the signal in the executed search stage just before the lower search And lower search process but the stored signal comprising a section similar to the lower search key signal, searches from the lower accumulation signal database
It is characterized by including .

Program of the present invention, as signal detection device described above, a signal detection program for causing a computer to function.

  The recording medium of the present invention is a computer-readable recording medium on which the above program is recorded.

  As described above, according to the present invention, in order to cut out the search key from coarse to dense step by step, compared to the conventional method that performs processing to cut out the search key signal comprehensively, It is possible to search for similar signals from the stored signal database using a smaller number of search key signals in the upper stage, and in the middle stage, the stored signals extracted from the stored signal database can be used to store the stored signals in the stored signal database. From the input signal, a stored signal having a section similar to the search key signal in the lower search stage is extracted, and the lower stored signal database (for the dense search in the embodiment) is extracted from the extracted stored signal. Low-order accumulated signal data with a smaller number of accumulated signals than the accumulated signal database. By searching for a stored signal similar to the search key signal separately cut out from the database, search processing is performed from a database with a smaller number of stored signals, which is faster than the conventional method. There is an advantage that signal detection processing can be performed.

Further, according to the present invention (second embodiment), in the upper search stage (primary search), when compared with the accumulated signal, the shift width of the upper search key signal for the upper search stage cut out from the input signal is increased. By doing so, it is possible to reduce the number of upper search keys to be compared with the accumulated signal stored in the accumulation database, and it is possible to search for accumulated signals having similar sections at higher speed.
Further, according to the present invention (second embodiment), in the lower search stage (secondary search), the information that identifies the accumulated signal searched in the upper search stage listed is used, accumulation signal to be stored in the storage signal database can narrow down (music in the second embodiment), the stored signal that compares with the lower search key signal for searching is reduced, the stored signal to be extracted, compared with the conventional It is possible to search at high speed.

  Further, according to the present invention (second embodiment), the search key signal signal (musical piece) used is large because the shift amount of the search key signal with respect to the accumulated signal is large in the upper search stage (primary search). Is not used in the exact interval (from what second to what second) of the accumulated signal (music) stored in the accumulated database, but in the lower search stage (secondary search), There is an advantage that a search with a fine shift width can be performed, and the interval of the accumulated signal corresponding to the interval of the used search key signal can be detected with fine information, that is, with high time accuracy.

  The present invention gradually searches for accumulated signals having a high detection accuracy in retrieval of accumulated signals having sections similar to the input signal from the accumulated information database in a plurality of stages, and is coarse in the upper level. A search key signal (for example, a search key signal for coarse search in the embodiment) is cut out from the input signal at intervals, and each search is performed from an accumulated signal database (for example, an accumulated signal database in the embodiment) with a small number of search key signals. Select a stored signal that is similar to the key signal for use, and use the selected stored signal to create a lower database that collects a smaller number of stored signals than the higher level used by the lower search unit, and compare this lower database to the higher level. Then, a lower search key signal (for example, a search key signal for fine search in the embodiment) cut out from the input signal at a dense interval is used. There are, databases for backward search stage (e.g., in embodiments dense search accumulation signal database) searches for a signal similar from.

In this way, by performing the low-order search stage multiple times, similar signals can be detected with high time accuracy by reducing the number of searches. The detection process can be performed at a higher speed than when the key signal is used.
That is, according to the present invention, there is an accumulated signal database in which a plurality of accumulated signals are registered in advance. Whether or not a signal similar to a certain section of the accumulated signal is included is searched for each section in which the input signal is cut out. At this time, the present invention divides the section while reducing the shift width of the section from the input signal for each search stage, and gradually narrows down the accumulated signals to be searched in a plurality of stages to obtain an input signal with high detection accuracy. A similar accumulated signal detection process is performed at high speed.
Here, the search at each stage is performed by comparing a predetermined section of the input signal with a part or all of the stored signals registered in advance.
Hereinafter, a signal detection device according to an embodiment of the present invention will be described with reference to the drawings.

<First Embodiment>
FIG. 1 is a block diagram showing a configuration example of the first embodiment. In the signal detection apparatus according to the first embodiment shown in FIG. 1, signal detection in a plurality of stages is described in two stages of a coarse search and a fine search. It goes without saying that the signal detection process can be performed in the same process even in two or more sub-stages, that is, three or more stages. 1 includes a rough search search key signal cutout unit 101 (upper search key signal cutout unit), a coarse search unit (upper search unit) 102, a dense search accumulated signal database creation unit (lower accumulated signal accumulated signal). Database creation unit) 103, search key signal cutout unit for fine search 104 (lower search key signal cutout unit), fine search unit 105 (lower search unit), accumulated signal database 111, and accumulated signal database for fine search 112 (lower accumulated signal) Database).

In the accumulated signal database 111, acoustic signals or video signals as a plurality of accumulated signals are registered in advance together with identification information for specifying each.
The coarse search search key signal cutout unit 101 cuts out a coarse search search key signal from a plurality of locations of an input signal input from the outside.
The coarse search unit 102 searches the accumulated signal database 111 for an accumulated signal including a section similar to the coarse search key signal, and uses the accumulated signal thus searched as an accumulated signal for fine search. The identification information to be specified is output to the dense search accumulated signal database creation unit 103 at the lower stage.
The fine search accumulated signal database creation unit 103 collects the fine search accumulated signals output from the coarse search unit 102, registers the accumulated signals in the database in correspondence with the identification numbers, and performs the fine search accumulated signal database. 112 is generated.

The fine search search key signal cutout unit 104 cuts out a fine search search key signal from a plurality of locations of an input signal as a reference signal input from the outside. Here, the cut-out interval of the search key signal for coarse search is larger than the cut-out interval of the search key signal for fine search (the cut-out shift width is long). Further, the coarse search search key signal cutout unit 101 may sequentially cut out the coarse search search key signal from cutout portions that are thinned out in time.
The dense coarse search unit 105 searches the dense search accumulated signal database 112 for an accumulated signal having a section similar to the fine search search key signal, and outputs a search result. From this search result, it is detected that the accumulated signal similar to the search key signal for fine search is included in the input signal at the extracted location at the location where the search key for fine search is extracted.

In the first embodiment of the present invention, an acoustic signal is described as an example of the input signal and the accumulated signal.
Hereinafter, the operation of the signal detection apparatus according to the first embodiment will be described with reference to FIGS. 1 and 2. FIG. 2 is a flowchart showing an operation example of the signal detection apparatus according to the first embodiment.
First, the coarse search search key signal cutout unit 101 receives an input signal from an external device, and designates a specified shift width (for example, a shift width of a rough search search key signal set or preset by the user). In the case of a feature vector code string to be described later, the search key signal length (for example, a search key signal length for coarse search set or set in advance by the user) In the case of a code string of (2), the search key signal for coarse search is sequentially cut out from the inputted input signal, and the search key signal for coarse search is output to the coarse search device 102. (Step S101).

Next, for each of the coarse search search key signals input from the coarse search search key signal cutout unit 101, the coarse search unit 102 obtains a similar accumulated signal for each coarse search key signal in the DAL-2. The stored signal database 111 is searched by a search method (described later).
When the rough search unit 102 detects that the similarity between the search key signal for coarse search and the interval of the accumulated signal is larger than a preset threshold value, the coarse search unit 102 selects one of the accumulated signals including the interval, Among the selected accumulated signals, the identification number (information for identifying the accumulated signal) of the accumulated signal having the top N similarities is output to the dense search accumulated signal database creating apparatus 103 (step S102).

  Next, the dense search accumulated signal database creation device 103 searches the accumulated signal database 111 based on the identification information input from the coarse search device 102 and reads the accumulated signal corresponding to the identification information from the accumulated signal database 111. The read accumulation signal is registered in the fine search accumulation signal database 112, and has a section similar to the coarse search search key signal, and is composed of N accumulation signals whose similarity exceeds a preset threshold value. Create a database. For example, if 3N accumulated signals are registered in the accumulated signal database 111, the number of accumulated signals to be searched in the dense search is N, and the search target is narrowed down to 1/3, and the time accuracy is fine. Effective search can be applied effectively. When the fine search accumulation signal database creation device 103 completes the creation of the fine search accumulation signal database 112, the fine search accumulation signal database creation completion output to the fine search search key signal extraction device 104 is output. (Step S103).

  When a fine search accumulation signal database creation completion notification is input from the fine search accumulation signal database creation apparatus 103, the fine search search key signal extraction apparatus 104 receives a specified shift width (for example, set by a user or previously set). The fine search is performed based on the fine search search key signal shift width (in units of bits) and the fine search search key signal length (for example, a fine search search key signal length set or set in advance by the user). The search key signal for search is sequentially cut out from the input signal, and the search key signal for fine search cut out is sequentially output to the fine search device 105 (step S104).

Next, for each fine search search key signal input from the fine search search key signal cutout unit 104, the fine search unit 105 stores each of the N accumulation signals registered in the fine search accumulation signal database 112. Compare with. Here, the fine search unit 105 reads the accumulation signal from the fine search accumulation signal database 112, and moves the fine search search key signal with respect to the read accumulation signal by a preset shift width. The stored signal having a section similar to the search key signal for fine search is searched. When a search signal having a similarity level exceeding a preset threshold is detected for each search key signal for fine search, the fine search unit 105 extracts the search key signal for fine search from the input signal. It is detected that the same signal as a similar section in the detected accumulated signal is included in the location, and the accumulated signal and the similar section are output as a detection result (step S105).
Here, the threshold value of the similarity in the dense search is not necessarily the same as that in the coarse search stage, and a higher similarity value or a lower similarity value than in the coarse search stage can be set. In particular, by setting a high similarity value as a threshold value, it is possible to expect a practical effect of reliably detecting a stored signal having high similarity as compared with a rough search, that is, selection of a candidate music piece.

The search method of DAL-2 described above is described in Patent Document 2, but in order to clarify the operation of this embodiment, an outline will be described below together with TAS and DAL-1 as a supplement.
TAS is a high-speed search method for audio signals and video signals. Feature vectors are extracted in time series from the target signal (corresponding to the input signal of this embodiment) and the accumulated signal (acoustic signal or video signal), respectively. As the accumulated signal, feature information is extracted in advance, and is registered in the accumulated signal database 111 as a code string by vector quantization described later in the format of the feature information. As the feature vector, a power spectrum can be used in the case of an audio signal, and a color feature vector for each frame can be used in the case of a video signal.

Next, the extracted feature vector is converted into a code string by vector quantization. After this conversion, a collation window having the same length as the target signal is set as the accumulated signal.
Then, the appearance frequency for each code is examined for the target signal and the code string of the accumulated signal in the verification window, and a histogram is created.
The degree of similarity between the target signal histogram created as described above and the stored signal histogram is measured, and whether or not the target signal has been detected is determined based on whether or not the measured value is equal to or greater than a set value. Is called. In this case, as a similarity measure, it is possible to use a histogram overlap rate or the like.

The above-described similarity detection process is repeatedly performed while sequentially shifting the position of the matching window on the accumulated signal in the time axis direction (in this embodiment, the search key signal is shifted with respect to the accumulated signal).
When the position of the matching window is shifted, it is not always necessary to perform one feature vector at a time. Instead, the movable width is obtained from the similarity at the current position and the set value of the similarity as a detection reference. It is also possible to shift several at a time. This movable width can be calculated from the viewpoint that the degree of similarity does not reach the set value no matter how many feature vectors are exchanged, and a specific calculation method is already known and will be omitted.

DAL-1 is a search method for searching for music flowing in the background even when foreground sound mainly having a line spectrum structure such as narration exists. For each of the target signal (corresponding to the input signal of this embodiment) and the accumulated signal, the acoustic spectrum is divided into one or more numbers in the time and frequency directions.
Next, search processing is performed by comparing each divided small region (search key signal in the input signal). As a search method used at this time, the above-described TAS can be used. And the search result with respect to each small area | region is integrated by a voting method, The location which many small areas have a high similarity is specified, and it searches.

  DAL-2 is a modification of the DAL-1 small region verification method. That is, DAL-2 vector-quantizes a small area in DAL-1, calculates the similarity between quantized codes in advance, creates a similarity table, and stores each accumulated signal (registered in the database). In contrast, the present invention is characterized in that a table in which the appearance position of each quantization code is checked in advance is created. By using these tables, almost equivalent search results can be obtained at a higher speed than DAL-1.

Next, the results of signal detection experiments performed by the signal detection apparatus according to the first embodiment described above will be described with reference to the drawings.
In this experiment, an acoustic signal of an experimental program is used as an input signal, and a stored signal database 111 that collects music acoustic signals as a stored signal is prepared, and the stored music used to detect the stored signal used by the input signal is used. Suppose that a detection experiment is performed. Then, this music use detection experiment was performed by comparing the method of the first embodiment and the conventional method on a computer, and the time required for processing by each method and the detection accuracy were calculated. Comparison was made between the invention and the conventional example. Two experimental programs, program A and program B, were prepared, and the above-described music use detection experiment was performed for each program.

At this time, in the first embodiment, the coarse search unit 103 selects the search key signal for search and the interval of the accumulation signal from the accumulation signals including similar intervals for each group search search key signal. The top three similarities are output to the dense search accumulated signal database creation unit 103.
The coarse search search key signal cutout unit 101 cuts out the coarse search search key signal from the input signal by setting the coarse search key signal shift width to 5 seconds and the coarse search search key signal length to 5 seconds. To do. The search key signal cutout unit 104 for fine search cuts out the search key signal for fine search from the input signal by setting the search key signal shift width for fine search to 1 second and the search key signal length for fine search to 5 seconds. It was.

Then, an experiment was performed in searching for an accumulated signal having a section similar to a certain section of the input signal from the accumulated signals registered in the accumulated signal database 111.
Further, in the search using DAL-2 in the dense search unit 105, in the extraction of the search key signal for fine search, the search is divided into twice as many small regions as the search by the DAL-2 in the coarse search unit 102. It was.
On the other hand, in the conventional method, the search key signal for search having a length of 5 seconds is cut out from the input signal by shifting it by 1 second, and the stored signal including a section similar to each search key signal for search is stored as the stored signal. We searched from the database and decided to output the result.
The DAL-2 parameter setting in the conventional method is the same as the DAL-2 parameter setting of the coarse search unit 102 of the first embodiment.

  The experimental results are shown in the tables in FIGS. The table of FIG. 3 is a detection speed comparison table showing the detection speed in the first embodiment when the detection speed of the conventional method is 1. From the table in FIG. 3, it can be seen that the first embodiment can detect the used music (accumulated signal) from the accumulated signal database at a speed four times faster than the conventional method. Here, the detection speed refers to a playlist (a table in which data for identifying the stored music used in the program, that is, the data to be detected) is listed after the input signal is input to the signal detection device. The search result after the input signal in the conventional method (when searching for all accumulated signals under the same search conditions as for the dense search of the present invention) is input to It is expressed as a ratio based on the reciprocal of the time until output.

Next, the table of FIG. 4 is a table comparing detection accuracy between the conventional method and the first embodiment.
Here, the correct answer rate is the proportion of the detected songs that are correct, and the detection rate is the proportion of the songs that are used in the input signal and that are detected from the songs that have the same accumulated signal in the accumulated signal database 111. .
And the table of FIG. 4 shows the detection rate and correct answer rate in the proposed method of the first embodiment when the correct answer rate and the detection rate of the conventional method are 1.

  In the first embodiment of the present invention, there are a case where the detection accuracy is improved and a case where the detection accuracy is lowered as compared with the conventional method. That is, as a case where the detection accuracy is improved, the number of music pieces extracted from the music stored in the accumulated signal database 111 by the coarse search key signal and registered in the dense search accumulated signal database 112 is compared. As a result, the false detection in the dense search unit 105 may be reduced. On the other hand, as a case where the detection accuracy is lowered, it is conceivable that due to the search omission of the rough search unit 102, the music used is missing at the stage of registration in the dense search accumulated signal database 112.

From the experimental results shown in the tables of FIG. 3 and FIG. 4 described above, it is considered that the detection accuracy of the stored information from the stored signal database in the embodiment of the present invention is almost equivalent to the conventional method. On the other hand, the detection speed in the embodiment of the present invention is four times or more that of the conventional method.
Therefore, the signal detection method in the embodiment of the present invention is considered to be effective as a faster signal detection method while maintaining the same detection accuracy as the conventional one.
Further, the present invention is considered to be particularly effective for detecting the use of music in a program in which the same music is repeatedly used, such as a TV drama.
In the experiment described above, the search key signal length for coarse search and the search key signal length for fine search are set to the same 5 seconds. However, the length is not limited to this, and the length is different from the length described above. The search key signal length for coarse search and the search key signal length for fine search may be set as different lengths.

In this embodiment, DAL-2 is used as the search method used for the coarse search and the fine search. However, TAS, BAM (binary area search method), or the like may be used for the search.
Different search methods may be used for the coarse search and the fine search.
Further, the search key signal extraction device 101 for rough search can be used only when starting the search key signal for rough search from a specified section or location (for example, set by the user) on the input signal. It is.

Further, the search key signal extraction device 104 for fine search detects the search key for coarse search detected by the rough search unit 102 that a certain accumulated signal section is similar with a degree of similarity larger than a preset threshold. It is also easy to cut out the search key signal for fine search only from the vicinity of the signal cut-out location.
In this case, the fine search search key signal extraction device 104 can reduce the number of fine search search key signals to be extracted, and can perform signal detection at higher speed than the conventional method.

  Further, in the first embodiment, the search signal for rough search is cut out only once, and the accumulated signal registered in the accumulation signal database for fine search by rough search and the accumulation signal database for fine search is reduced only once. The coarse search accumulated signal database thus obtained is used as an accumulated signal database to perform rough search search key signal extraction, coarse search and fine search accumulated signal database creation, The configuration may be such that the reduction of stored signals registered in the stored signal database is repeated, that is, performed in a plurality of stages, while changing the parameters and the search method in the process of creating the stored signal database for coarse search and fine search.

For example, as a configuration, the coarse search search key signal cutout unit 101 cuts out the coarse search search key signal from the input signal, and the coarse search unit 102 extracts the coarse search search key signal from the accumulated signal database 111 as described above. A stored signal having a section similar to is searched.
Next, the dense search accumulated signal database creation unit 103 creates the dense search accumulated signal database 112 in which the searched accumulated signals are registered.
Next, the fine search search key signal cutout unit 104 cuts out the fine search search key signal from the input signal.
When the search key signal for fine search is extracted, the fine search unit 105 searches the accumulation database 112 for fine search for an accumulation signal similar to the search key signal for fine search.

Then, the dense search accumulated signal database creation unit 103 newly re-registers the accumulated signal searched by the dense search unit 105 in the immediately preceding dense search stage in the fine search database 112.
Next, the fine search search key signal cutout unit 104 cuts out the fine search search key signal from the input signal. At this time, the dense search search key signal cutout unit 104 further reduces the cutout shift width and improves the time accuracy in the search as compared with the cutout in the immediately preceding search stage (the search stage higher than itself). .
Then, the dense search unit 105 searches for an accumulation signal similar to the search key signal for fine search from the accumulation database 112 for fine search re-registered to a number further reduced as compared to immediately before.

That is, in the upper-level search, the search signal for rough search is cut out from the input signal at rough intervals, and the stored signal having a certain degree of similarity from the stored signal database 111 by a small number of search key signals for rough search. (For example, as described above, an accumulated signal having a detected degree and a similarity exceeding a preset threshold value) An accumulation signal having a certain degree of similarity in the search key signal for fine search extracted from the input signal at a fine interval from the accumulated signal database 112 compared to the case of the immediately preceding upper search stage (for example, the above-described upper signal) Compared to the search stage, the stored signal having a detected similarity and a degree of similarity exceeding a preset threshold as a higher numerical value) is searched. By repeating this process, the search can be performed with gradually increasing detection accuracy at each stage.
For example, when the search process is performed in a plurality of stages of three or more stages, the shortest signal of the search key signal with respect to the input signal in each stage from the highest coarse search stage to the lowest dense search stage. The length (collation window length), the time movement width of the search key signal (movement width of the collation window) are refined (reduced), and the stored signal is narrowed down step by step.

  At this time, the search key signal extraction unit 104 for fine search extracts the search key signal for fine search from the input signal. As described above, the search key signal extraction unit 104 is preset for each search stage (or designated by the user). The search key signal for fine search is cut out at the cutout interval (shift width). Here, the shift width is set larger in the upper search process than in the lower search process. That is, the shift width in cutting out the search key signal for coarse search is larger than that for the search key signal for fine search. Further, the search key signal for fine search becomes smaller as compared with the shift width in the cut-out of the search key signal for the previous fine search as it goes to the lower-level search process.

As described above, the signal detection apparatus according to the first embodiment performs the lower-level search in the dense search accumulated signal database creation unit 103, the fine search search key signal cutout unit 104, and the fine search unit 105 a plurality of times, The detection accuracy in the stored information search may be gradually improved.
In addition, a plurality of configurations corresponding to the dense search accumulated signal database creation unit 103, the fine search search key signal cutout unit 104, and the fine search unit 105 that perform search processing at each search stage may be provided.

As described above, the signal detection apparatus according to the present embodiment includes the accumulated signal used as the input signal from the accumulated signals registered in the accumulated signal database 111 in the two stages of the coarse search and the fine search. Search processing is speeded up by narrowing down the number of accumulated signals to be searched.
On the other hand, methods that improve the efficiency of pattern recognition and signal search processing by sequentially performing coarse search processing and dense search processing are names such as coarse / fine or coarse to fine. Is widely known.

However, the present invention is not intended to simply apply a known density method to signal detection. This will be described below using an example of detecting a song being broadcast.
First, in the known coarse / fine method, in the input signal (broadcast acoustic signal), a collation window is set for the accumulated signal with a coarse movement width, and fine movement near the location where the music to be detected is detected. The collation window is set by the width, and the detection time, that is, the detection position in the accumulated signal is specified in detail.
According to such a method, it is not always possible to obtain the same result as compared with the method of matching with a small movement width from the beginning, but it is expected to reduce the number of times of matching and improve processing efficiency. can do.

  On the other hand, as already described in detail, in the present embodiment of the present invention, a dense search accumulated signal database 112 which is a set of accumulated signals extracted by a coarse search process and having a higher similarity is newly created. That is, the accumulated signal (target music piece) detected as having a high similarity in the collation with a rough movement width is selected, and the search is performed using only the selected accumulation signal in the collation with a fine movement width. It is a feature.

Collation with only a small movement width by the conventional method requires a large amount of calculation. For this reason, as described above, by performing a dense search using only a small number of selected accumulated signals (target music pieces), the signal processing apparatus of this embodiment reduces the processing cost in signal processing by the conventional method. It can be significantly reduced compared to
As described above, the present embodiment selects a stored signal to be subjected to a fine search from the stored signal database 111 through a rough process, and obtains a final search result from the selected stored signal through a dense search process. It has become.
In particular, when searching for music for a broadcast program or the like, specific music is often used repeatedly for the purpose of providing consistency in the program, and this embodiment functions extremely effectively.

Note that the search method used at each stage is not necessarily the same algorithm, and the search parameter setting can be changed. For example, in the example described here, the small areas are arranged so as to have a larger number in the dense search than in the coarse search, that is, at a high density. Other search parameters include a similarity threshold, a frequency band, a search key signal length, and the like.
Further, in the present embodiment, the input signal input from the outside and the accumulated signal registered in the accumulated signal database 111 are extracted in a time series by extracting feature information from the sound signal or the image signal data itself. Any of the code strings may be used.
That is, as already described in this embodiment, feature information may be extracted from the data of an acoustic signal or an image signal, and similarity may be obtained from the feature information. Or any other method may be used. Hereinafter, although an input signal is described as an acoustic signal, the same processing can be performed on other signals such as an image signal.

Further, when similarity is obtained from feature information, the input signal is already feature information, and the stored information registered in the stored signal database 111 is also feature information. The stored information registered in the database 111 is feature information. When obtaining similarity, the feature information is extracted from the input signal and compared, and the input signal is the feature information of the acoustic signal. When the stored information registered in 111 is the acoustic signal itself and the similarity is obtained, the feature information is extracted from the stored information for comparison, and the input signal is the acoustic signal itself and registered in the accumulated signal database 111. When the stored information is the acoustic signal itself and the similarity is obtained, the feature information is extracted from the input signal and the stored information for comparison. May be used either configuration.
Further, in the case of a technique for calculating a correlation between data itself and obtaining similarity, data of an image signal and an acoustic signal itself are registered in the accumulated signal database 111.

<Second Embodiment>
Next, a second embodiment of the present invention will be described with reference to FIG. FIG. 5 is a block diagram illustrating a configuration example of the signal detection apparatus according to the present embodiment. Here, the signal detection device is specifically realized on a general computer having a CPU and a memory. The same reference numerals are given to the same configurations in the first embodiment, and different configurations and different functions will be described. The search method in the coarse search unit 102 and the dense search unit 105 in the second embodiment is the same as that in the first embodiment, but a feature extraction unit is provided to clarify the feature information extraction process. .

The feature extraction unit 106 extracts feature information from the input audio signal or video signal and outputs it as a target feature signal.
In the original database 110, characteristic information of a plurality of audio signals or video signals is registered as accumulated signals.
The accumulated signal database 111 stores an accumulated signal selected from the original database 110 by the rough search unit 102 corresponding to the input signal.

As in the first embodiment, the coarse search search key signal cutout unit 101 sequentially performs coarse search from the target feature signal while shifting the cutout position according to the set shift width of the coarse search search key signal. The search key signal is cut out.
The music list creation unit 107 is information that specifies the stored signal searched by the rough search unit 102 using the search key signal for rough search, for example, if the stored signal is an acoustic signal, a song name, an identification number, etc., or a video signal. If so, list the video name, identification number, etc. in the coarse search list.

The fine search accumulated signal database creation unit 103 refers to the coarse search list, reads the accumulated signals listed in the coarse search list from the accumulation database 111, and stores them in the fine search accumulated signal database 112. Then, a process for creating the dense search accumulated signal database 112 used for the fine search is performed.
As in the first embodiment, the fine search key signal cutout unit 104 has a set shift width of the fine search search key signal (a cut shift width is smaller than that of the coarse search search key signal). The search key signal for fine search is sequentially cut out from the reference signal while shifting the cut-out position.
The music list creation unit 108 is information that specifies the accumulated signal searched by the dense search unit 105 using the above-described secret search key signal. For example, if the accumulated signal is an audio signal, the music list creation unit 108 may be a song name, an identification number, For example, video names and identification numbers are listed in the fine search list.

Next, the signal detection apparatus according to the second embodiment will be described with reference to FIGS. 5, 6, and 7. Here, FIG. 6 is a conceptual diagram illustrating the extraction of search key signals for coarse search and fine search. FIG. 7 is a flowchart illustrating an example of signal detection operation according to the second embodiment.
In the present embodiment, an input signal is assumed to be a video signal or an audio signal that is broadcast or the like, or a recorded video signal or a recorded audio signal, and an accumulated signal (stored in the original database 110). As the stored signal), a signal obtained by converting video recorded in a video or the like or music recorded on a CD into feature information is assumed.

The signal detection device inputs a video signal or an audio signal that is broadcast from an external device or the like, or a recorded video signal or a recorded audio signal as an input signal, and inputs the input signal to the feature information extraction unit 106. Output (step S201).
When the input signal is input, the feature information extraction unit 106 extracts feature information based on the input signal, creates a target feature signal, and roughly searches the target feature signal via the management unit 109. The search key signal cut-out unit 101 is output (step S202).

Next, the coarse search search key signal cutout unit 101 cuts out the coarse search search key signal from the input target feature signal while shifting the cutout position by the set shift width, and sequentially searches the coarse search unit 102. (Step S203). Here, the coarse search search key signal cutout unit 101 performs a coarse search search key signal while shifting by the time movement width t1 according to the shortest signal length L of the video signal or audio signal for which the transmitted target feature information is to be found. Is cut out (see FIG. 6A).
The search key signal cutout unit 101 for coarse search reads an accumulated signal (characteristic information of a video signal or an audio signal) corresponding to an input signal, which is compared with the target feature signal, from the original database 110 and stores the accumulated signal database 111. (Step S204). Here, the accumulated signal is read from the original database 110 and registered in the accumulated signal database 111 is completed by the time of the first rough search, and the accumulated signal from the original database 110 is searched for each of the following lower-level search processes. No removal is performed.

When the search key signal for coarse search is input, the coarse search unit 102 compares the divided search key signal for coarse search, that is, the extracted search signal with the accumulated signal read from the accumulated signal database 111, and compares the comparison. As a result, the similarity is calculated and output (step S205).
The music list creation unit 107 inputs the similarity of each of the compared accumulated signals, compares it with a preset threshold value, and determines that the accumulated signal having the similarity exceeding the threshold value is a match in the rough search process. Then, the data (music name, video name, identification information, etc.) specifying the accumulated signal determined to be coincident is listed to generate a rough search list, and this rough search list is output to the management unit 109 (step S206).

Next, the management unit 9 refers to the input coarse search list, extracts an accumulated signal from the accumulated signal database 111 using data specifying the accumulated signal in the coarse search list, and stores the accumulated signal described in the coarse search list. Signals are collected, and the target feature signal is output (transmitted) to the fine signal search key signal, and the coarse search list and the collected accumulated signal are output to the fine search accumulated signal database creation unit 103 (step S207).
When the coarse search list and the collected accumulated signal are input (received), the fine search accumulated signal database creation unit 103 registers the coarse search list and the inputted accumulated signal in the database, and the fine search accumulated signal database 112. Is created (constructed) as a database storing a minimum accumulated signal for dense search (step S208).

Next, as shown in FIG. 6B, the fine search search key signal cutout unit 104 selects the target feature signal that is the feature of the video or audio signal to be found in the same manner as in the coarse search process. The search key signal for fine search is cut out from the target feature signal while being shifted by the shortest signal length L and by a shorter time movement width t2 (<t1) in the search process of the coarse search, and sequentially output to the fine search unit 105. (Step S209).
When the search key signal for fine search is input, the fine search unit 105 reads and acquires an accumulation signal to be compared from the accumulation signal database 112 for fine search (step S210).

Then, the fine search unit 105 compares the input search key signal for coarse search with the accumulated signal read from the accumulated signal database 111, and calculates and outputs the similarity as the comparison result (step S211). .
The music list creation unit 212 receives the similarity of each accumulated signal compared from the dense search unit 105, compares it with a preset threshold value, and uses the accumulated signal having the similarity exceeding this threshold value as a dense search process. The data (music name, video name, identification information, etc.) specifying the stored signal determined to match is listed up, a dense search list is generated, and displayed on the screen via the management unit 109 A notification process to the user is performed (step S212).

  Note that a program for realizing the function of the signal detection process in the flowcharts of FIGS. 2 and 7 is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read into a computer system and executed. Thus, signal detection processing may be performed. The “computer system” here includes an OS and hardware such as peripheral devices. The “computer system” includes a WWW system having a homepage providing environment (or display environment). The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Further, the “computer-readable recording medium” refers to a volatile memory (RAM) in a computer system that becomes a server or a client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. In addition, those holding programs for a certain period of time are also included.

  The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

It is a block diagram which shows the structural example of the signal detection apparatus by the 1st Embodiment of this invention. It is a flowchart which shows the operation example of the signal detection apparatus by 1st Embodiment of FIG. It is the table which showed the result of the comparison experiment of the detection speed of 1st Embodiment and the conventional method. It is the table which showed the result of the comparison experiment of the detection accuracy of 1st Embodiment and the conventional method. It is a block diagram which shows the structural example of the signal detection apparatus by the 2nd Embodiment of this invention. It is a conceptual diagram explaining the extraction of the search key signal for rough search and fine search from the input signal in the second embodiment. It is a flowchart which shows the operation example of the signal detection apparatus by 2nd Embodiment of FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 101 ... Rough search search key signal cutout part 102 ... Rough search part 103 ... Dense search accumulation signal database search part 104 ... Fine search search key signal cutout part 105 ... Dense search part 106 ... Feature extraction part 107, 108 ... Music List creation unit 109 ... management unit 110 ... original database 111 ... accumulation signal database 112 ... accumulation signal accumulation signal database

Claims (4)

A signal detection device that compares a predetermined section of an input signal with a part or all of a pre-registered accumulated signal and detects accumulated signals similar to the input signal at a plurality of search stages. ,
An accumulated signal database in which accumulated signals are registered;
An upper search key signal cutout unit that cuts out the upper search key signal from the input signal at a predetermined interval ;
An upper search unit that searches the stored signal database for an accumulated signal including a section similar to the upper search key signal, and outputs the searched accumulated signal;
And the lower accumulation signal database creation unit that creates a lower accumulation signal database a collection of accumulated signals the upper search unit has output,
Extraction of upper search key signal in which lower search key signal is searched at an interval narrower than the interval in upper search key signal extraction unit, and the stored signal including a similar section in the upper search unit is searched among the input signals. A low-order search key signal cutout unit that cuts out only from the area near the location ;
A signal detection apparatus comprising: a lower search unit that searches an accumulation signal including a section similar to the lower search key signal from the lower accumulation signal database. This is a signal detection method in which a predetermined interval of an input signal is compared with an interval of an accumulated signal registered in advance in an accumulated signal database, and an accumulated signal similar to the input signal is detected in a plurality of search stages. And
The plurality of search stages includes one coarse search stage and one or more dense search stages that are sequentially executed following the coarse search stage.
The rough search step includes:
An upper search key signal cutout process in which the upper search key signal cutout unit cuts out the upper search key signal from a plurality of locations of the input signal at predetermined intervals ;
An upper search process in which an upper search unit searches the stored signal database for an accumulated signal including a section similar to the upper search key signal, and outputs the searched accumulated signal;
Including
The dense search step includes:
A lower accumulated signal database creating unit for creating a lower accumulated signal database in which accumulated signals searched by the search of the search stage executed immediately before the dense search stage are created by the lower accumulated signal database creating unit;
If the lower search key signal cut-out unit searches for the lower search key signal , and the search stage executed immediately before of the input signals is a coarse search stage, stored signals including similar sections are searched for in the coarse search stage. If the search stage executed immediately before is only from the area near the cut-out location of the upper search key signal, and if the search stage executed immediately before is a dense search stage, an accumulated signal including a similar section was searched in the previous dense search stage. A low-order search key signal cut-out process that cuts out only in the vicinity of the cut-out position of the low-order search key signal at an interval narrower than the signal cut-out interval at the search stage executed immediately before the dense search stage ;
A lower search process in which a lower search unit searches the lower stored signal database for an accumulated signal including a section similar to the lower search key signal;
Signal detection method, which comprises a.   A signal detection program for causing a computer to function as the signal detection device according to claim 1. A computer-readable recording medium on which the program according to claim 3 is recorded.

Images